Of Models and Machines: Implementing Bounded Rationality

摘要

This essay explores the early history of Herbert Simon's principle of bounded rationality in the context of his Artificial Intelligence research in the mid 1950s. It focuses in particular on how Simon and his colleagues at the RAND Corporation translated a model of human reasoning into a computer program, the Logic Theory Machine. They were motivated by a belief that computers and minds were the same kind of thing-namely, information-processing systems. The Logic Theory Machine program was a model of how people solved problems in elementary mathematical logic. However, in making this model actually run on their 1950s computer, the JOHNNIAC, Simon and his colleagues had to navigate many obstacles and material constraints quite foreign to the human experience of logic. They crafted new tools and engaged in new practices that accommodated the affordances of their machine, rather than reflecting the character of human cognition and its bounds. The essay argues that tracking this implementation effort shows that "internal" cognitive practices and "external" tools and materials are not so easily separated as they are in Simon's principle of bounded rationality-the latter often shaping the dynamics of the former.